Mitigation Policies for the Paris Agreement: An Assessment for G20 Countries
  • 1 0000000404811396 Monetary Fund
  • | 2 0000000404811396 Monetary Fund

Contributor Notes

Following submission of greenhouse gas (GHG) mitigation commitments or pledges (by 190 countries) for the 2015 Paris Agreement, policymakers are considering specific actions for their implementation. To help guide policy, it is helpful to have a quantitative framework for understanding: i) the main impacts (on GHGs, fiscal balances, the domestic environment, economic welfare, and distributional incidence) of emissions pricing; ii) trade-offs between pricing and other (commonly used) mitigation instruments; and iii) why/to what extent needed policies and their impacts differ across countries. This paper provides an illustrative sense of this information for G20 member countries (which account for about 80 percent of global emissions) under plausible (though inevitably uncertain) projections for future fuel use and price responsiveness. Quantitative results underscore the generally strong case for (comprehensive) pricing over other instruments, its small net costs or often net benefits (when domestic environmental gains are considered), but also the potentially wide dispersion (and hence inefficiency) in emissions prices implied by countries’ mitigation commitments.


Following submission of greenhouse gas (GHG) mitigation commitments or pledges (by 190 countries) for the 2015 Paris Agreement, policymakers are considering specific actions for their implementation. To help guide policy, it is helpful to have a quantitative framework for understanding: i) the main impacts (on GHGs, fiscal balances, the domestic environment, economic welfare, and distributional incidence) of emissions pricing; ii) trade-offs between pricing and other (commonly used) mitigation instruments; and iii) why/to what extent needed policies and their impacts differ across countries. This paper provides an illustrative sense of this information for G20 member countries (which account for about 80 percent of global emissions) under plausible (though inevitably uncertain) projections for future fuel use and price responsiveness. Quantitative results underscore the generally strong case for (comprehensive) pricing over other instruments, its small net costs or often net benefits (when domestic environmental gains are considered), but also the potentially wide dispersion (and hence inefficiency) in emissions prices implied by countries’ mitigation commitments.

I. Introduction and Summary

The landmark 2015 Paris Agreement lays the foundations for meaningful action to begin slowing atmospheric accumulation of heat-trapping gases. The centerpiece of the Agreement is commitments by 190 countries to reduce greenhouse gas (GHG) emissions, as specified in their ‘nationally determined contributions’ (NDCs).1 While these contributions are not consistent with containing mean projected warming to 2°C,2 are partially dependent (for some countries) on external finance, and are not legally binding3, countries are expected to report progress on meeting their NDCs (every two years starting 2018) and to submit updated NDCs (every five years). For G20 countries specifically—which collectively accounted for 81 percent of global carbon dioxide (CO2) emissions in 20154—NDCs target emissions (or, for China and India, emissions relative to GDP) mostly for 2030, though pledges differ in nominal stringency and baseline years against which targets apply (Table 1).

Table 1.

Paris Mitigation Pledges, Emissions Intensity and Emissions Per Capita, 2014

article image
Note: BAU denotes business as usual with no new mitigation measures and CO2 is from fossil fuels. Some countries (Argentina, Indonesia, Mexico) specify both conditional (contingent on external finance) and unconditional (not contingent) pledges—in these cases the unconditional pledges are included above.Source: and

It is long established in theoretical literature5 and increasingly appreciated among policymakers, international organizations, businesses, and others6 that carbon pricing would ideally be a key element of (energy-related) CO2 mitigation strategies7for two reasons. First, comprehensive carbon pricing provides across-the-board incentives for reducing energy use and shifting to cleaner fuels. Second, (just like existing fuel taxes) carbon pricing raises significant revenues: an important concern for finance ministries, as this allows the reduction of other burdensome taxes, funding of growth-enhancing investment, or alleviation of fiscal pressures. There is not a one-size-fits-all approach to mitigation, however: carbon pricing may be difficult politically in some countries, not least because of its first-order impact on energy prices, which underscores the importance of understanding trade-offs with other mitigation instruments.

To move policy forward, it is helpful to have quantitative assessments of the CO2, fiscal, local health and environmental, economic, and incidence impacts (across households and firms) of carbon pricing at the country level and of trade-offs among mitigation instruments: carbon taxes, emissions trading systems (ETSs), taxes on individual fuels, energy efficiency incentives, and so on. This information helps policymakers: choose among instruments; design policy specifics (e.g., policy stringency, use of potential revenues, compensation for vulnerable groups); and communicate the case for policy reform. At the international level, assessing whether countries are likely to fall short, meet, or exceed their targets under (explicit or implicit) carbon price scenarios informs dialogue on future NDC revisions and the possible need for coordination mechanisms.

This paper uses a reduced form model8 to provide an illustrative sense of this information for G20 countries. The model projects country-level fuel use by sector in a business-as-usual (BAU) scenario, assuming (for transparency of comparisons) no new mitigation policies. The environmental, fiscal, and economic welfare impacts of carbon pricing and other mitigation instruments are then computed, with all policies (again for transparency) scaled to impose the same explicit or implicit CO2 price,9 and using assumptions for fuel price elasticities, emission rates, and local externalities (e.g., air pollution mortality). Incidence analysis is also conducted for Canada, China, India, and the United States10 by linking the model to input-output tables and household expenditure surveys.

A large number of other (far more detailed) models have been used at the individual country (especially US), regional, and global level to assess the emissions implications of pricing CO2 (and sometimes other GHGs).11 Our approach is meant to complement this large body of literature in several respects, by: providing consistent comparisons across a broad range of countries12; evaluating policies against a broad range of metrics (emissions, fiscal, health, economic welfare and incidence)13; comparing pricing with a range of alternative mitigation instruments;14 and providing transparent explanations for why results differ across countries, policies, and parameter scenarios. Our reduced-form model is highly simplified, but it is parameterized such that it approximates projections and underlying behavioral responses for fuel use and emissions generated by more disaggregated, structural models (e.g., that account for detail on emerging technologies or feedback effects on fuel use from general equilibrium effects). Caveats to the analysis are extensively discussed below (see section II.B.), and are important to bear in mind, though in many cases they mainly involve choices over parameter values, and the implications of alternative assumptions are discussed through a sensitivity analysis (see section III.D.).

Some quantitative findings are summarized as follows.

Under a carbon tax alone, assumed here to have full coverage (i.e., the tax is applied to the carbon content of fossil fuel supply) and reaching, for illustration, a price of $70 per ton15 of CO2 by 2030, seven countries meet or exceed the CO2 component targets implied by their NDCs (Argentina, China, India, Indonesia, Russia, South Africa, and Turkey), another six countries would need further emissions reductions of up to 10 percent (Brazil, Germany, Japan, Korea, Mexico, and United States), and six countries would need further emissions reductions of above 10 percent (Australia, Canada, France, Italy, Saudi Arabia, and United Kingdom). These cross-country differences reflect differences in both the stringency of NDCs and in the relative price-responsiveness of emissions. For the whole G20, however, a carbon price of $35 per ton in 2030 is consistent with the total of current NDCs (given the disproportionate influence of China and India).

Carbon taxes can also raise substantial revenues, around 1–2.5 percent of GDP in most cases for the $70 per ton tax in 2030, and considerably more than that in a few cases. They also reduce deaths from local air pollution due to fuel combustion by roughly the same proportion as the reduction in CO2 emissions—the value of the reduction in local pollution deaths per ton of CO2 reduced in China, for example, is estimated at a striking $100. The pure welfare costs of the $70 per ton tax are generally not too large (less than 0.8 percent of GDP in all but three cases). However, accounting for local environmental benefits (but not global warming), the net welfare impacts are around zero to strongly positive in all but three cases.16

Coal taxes can be especially effective environmentally, achieving over 75 percent of the CO2 reductions under the carbon tax in six countries. For the same price and coverage, an ETS would have equivalent emissions impacts as the carbon tax in this analysis—hence the ‘carbon tax’ policy can also be interpreted as a fully comprehensive ETS. However, following typical practice to date (and to distinguish it analytically from the carbon tax), ETSs are assumed to cover large stationary emission sources only and as a result reduce emissions by between about 40 and 75 percent of the reductions under the (equally priced) carbon tax in most countries. A tax on CO2 emissions from power generation typically achieves 80 percent or more of the emissions reductions under ETSs. Energy efficiency policies, even if implemented nationwide (which would be practically challenging), typically reduce emissions by between 25 and 40 percent of that under the carbon tax. Road fuel taxes are a relatively weak instrument, typically reducing emissions by less than 10 percent of that under the carbon tax. Taxes on electricity output, rather than CO2 emissions from generating electricity, are also a weak instrument in countries with ample opportunities for switching to low-emission generation fuels.

Potential revenues from other instruments are generally well below those from carbon taxes—around 40–80 percent lower for (limited scope) ETSs (even with allowance auctions) and power sector taxes, and more than 70 percent lower for road fuel taxes and (China, India, and South Africa aside) coal taxes. The performance of other carbon mitigation policies at reducing air pollution deaths, relative to the carbon tax, follows a similar pattern (with some exceptions) to that of the relative reductions in CO2 emissions. Finally, the economic welfare comparison of other policy instruments can be largely anticipated. For instance, the welfare gains from coal taxes in China and India are close to those from the carbon tax, while those for road fuels are less than one-seventh of those for the carbon tax.

In short, the quantitative results, while illustrative, underpin three important (though qualitatively not especially novel) themes about carbon pricing. First is the generally strong environmental, fiscal, and economic welfare case for carbon taxes (or similar pricing instruments) as well as their positive domestic welfare gains (or at least zero/generally modest net costs) relative to other instruments. This supports the central role of carbon pricing in mitigation and that it can be in countries’ own national interests to move ahead unilaterally. Second, however, the high (and politically challenging) carbon prices implied by some NDCs underscore the importance of complementary measures to compensate vulnerable groups and facilitate clean technology investment (potentially enhancing the behavioral responses to carbon pricing). Third, the large cross-country dispersion in prices implied by current NDCs may need to be partially addressed in future revisions to NDCs and through complementary international mechanisms (if applicable), like price floor arrangements.

The rest of this paper is organized as follows. Section II outlines the model and related assumptions. Section III presents the main results and sensitivity/incidence analyses. Section IV elaborates on the practicalities of moving policy forward and concludes.

II. Methodology

This section briefly overviews the model, caveats, data sources, and policy scenarios, with details on the model equations and parameter justifications provided in Appendices 1 and 2.

A. Analytical Framework

Five fossil fuels are distinguished, namely coal, natural gas, gasoline, road diesel, and other oil products (used in power generation, petrochemicals, home heating, non-road vehicles, etc.). The model projects, out to 2030, annual use of fossil and non-fossil fuel use in three sectors—power generation, road transport, and an ‘other energy’ sector, where the latter represents an aggregation of direct energy use by households, firms, and non-road transport.

(i) Power sector

In the power sector, electricity demand (aggregated over intermediate and final users) in the BAU increases over time with growth in GDP, according to the income elasticity17 for electricity, and changes in electricity prices, which are a function of associated fuel supply cost projections. Higher electricity prices reduce electricity demand through promoting both energy efficiency and less use of electricity-consuming products/capital. Energy efficiency also improves autonomously over time with technological progress (e.g., due to the natural retirement of older, less efficient capital).

Power is potentially generated from coal, natural gas, oil, nuclear, biomass, hydro, and other renewables like solar and wind (not all G20 countries use all seven fuels) where higher unit generation costs for one fuel cause switching to other fuels (in proportion to their generation shares). Autonomous technological progress, assumed to be fastest for renewables, gradually reduces unit generation costs for each fuel over time. Changes in electricity demand induce proportional changes in generation from each fuel, and, in turn, electricity prices reflect a share-weighted average of unit generation costs.

(ii) Road transport

The road transport sector distinguishes gasoline (light-duty) vehicles and diesel (heavy- and, in some cases, light-duty) vehicles, though implicitly an increase in average on-road, fleet fuel economy can represent a shift towards hybrid and electric vehicles as well as higher efficiency of petroleum vehicles. Again, future fuel use varies positively with GDP (through income elasticities for vehicle use), negatively with fuel prices (through shifting to more fuel-efficient vehicles and less driving), and negatively with autonomous improvements in fuel economy.

(iii) Other energy

The other energy sector disaggregates small fuel users (households, low-emitting firms) from large (industrial) users (e.g., steel, aluminum, cement, refining, chemicals, construction, domestic aviation) to distinguish ETSs which often (e.g., in the EU) only cover the latter.18 Fuels potentially used by the other energy sector include coal, natural gas, oil products, biomass (e.g., for home cooking and lighting), and other renewables. Future fuel use varies positively with GDP (through income elasticities) and negatively with fuel prices (through changes in energy efficiency and product usage) and autonomous improvements in energy efficiency.

(iv) Externalities

Fossil fuel use produces CO2 emissions, according to the carbon content of fuels. It also causes premature deaths from exposure to air pollution (equal to deaths per unit of fuel use times fuel consumption, where the death rate is generally trending down over time as air pollution emission rates per unit of fuel use fall with greater deployment of control technologies).19 In addition, road fuels are implicitly associated with externalities from vehicle use (congestion, accidents, and road damage), which feed into the calculations of welfare impacts.20

B. Caveats

BAU is defined with no new mitigation policies beyond those already implicit in recently observed energy use and prices and therefore excludes, for example, future policies for faster deployment of renewables or raising of carbon prices or energy taxes. This is a common approach in developing BAU scenarios,21 avoids (somewhat speculative) assumptions on future policy specifics and provides clean comparisons of alternative mitigation instruments to the BAU and across countries.22 More detailed modelling of other prospective policies would be needed at the national level however, as individual countries tailor their own, idiosyncratic policy packages to implement NDCs.

While the assumed fuel price responses (see Section C) seem reasonable for modest fuel price changes, they may not be for dramatic price changes that might drive major technological advances, or non-linear adoption of technologies like carbon capture and storage (CCS), and even the potential growth of well-established technologies may be highly uncertain.23 For this reason, and questions over the political viability of high prices—the current global average CO2 price is only about $1 per ton24—carbon prices above $70 per ton are not considered.25

The three energy sectors in the model are de-coupled from one another and from the broader economy. The former assumption is becoming questionable as penetration of electric vehicles adds to electricity demand, though only modestly (in proportionate terms) for the foreseeable future. The latter assumption seems reasonable, for the most part, because the assumed price responsiveness of fuel use is broadly consistent with that implied by econometric evidence (Appendix 2) and (as noted below) results from macro-energy models which do account for these linkages. The main concern in this regard is the welfare calculations which exclude (potentially important) linkages with the broader fiscal system (see below).

A further caveat is the focus on longer term impacts (of anticipated, gradually phased, policies)—the model is effectively comparative static with fuel use responding instantly and fully to fuel price changes. Also of interest (but beyond our scope) is the shorter term, in which fuel price responsiveness is more muted—these dynamic responses might be explored, for example, with computable models incorporating gradual turnover of capital stocks. The model also assumes perfectly elastic fuel supply curves (which shift up in response to new taxes) and does not account for international trade (given the focus of NDCs on emissions from domestic fuel consumption). If major economies move forward simultaneously with significant mitigation measures, there will be some reduction in international fuel prices, an offsetting increase in fuel demand, and terms of trade effects, but again these impacts (which are sensitive to assumed policy scenarios and modelling of international fuel markets) are beyond our scope.

C. Parameterization

2014 fuel use for each G20 country is taken from IEA (2017b)26, aggregated to ensure consistency with the model’s energy sectors, and projected forward for the BAU using relationships discussed above (with mathematical specifics detailed in Appendix 1). Projected GDP is from IMF (2017) (assuming growth continues after 2022 at the same rate as projected for 2022) and rates of autonomous technological change are based on typical modelling assumptions.

Fuel taxes (or subsidies) by country are based on estimates using the price-gap approach—that is, the difference between (publicly available) retail prices and supply prices—and implicitly reflect the combined effect of excises, any favorable general sales tax (GST) provisions for household fuels, regulated or monopoly price distortions, and carbon pricing. Supply prices (or more precisely opportunity costs) are international reference prices for finished fuel products, with adjustments for transport and distribution margins, and are projected forward by averaging over (rising) international price projections in EIA (2018) and (generally flat) projections in IMF (2017) yielding real prices for coal and oil that are 31 and 42 percent higher in 2030 compared with 201527, and between about 0 and 30 percent higher for natural gas in different regions (Table 1, Appendix 2). BAU emissions projections, and the percent emissions reductions from carbon pricing, are both increased with lower fuel price projections (the latter because carbon pricing has a greater proportionate impact on fuel prices).

Income elasticities for energy products (between 0.5 and 0.8) are based on empirical evidence (see Appendix 2), with some country-specific adjustments (e.g., lower for China as a way of reflecting re-balancing away from energy-intensive activity, and higher for India to reflect expanding grid access and vehicle ownership). Fossil fuel use projections would be broadly in line with those in IEA (2017a) for the same energy price projections in the seven G20 countries they report; however, our lower price projections lead to significantly higher BAU fuel use and CO2 emissions (Table 2, Appendix 2), implying higher CO2 prices may be needed to meet mitigation pledges.

Table 2.

Impacts of Other Policies Relative to Carbon Tax, 2030

article image
Source: See text and Appendix.Note: aDue to rounding, slight reductions in revenues from the energy efficiency combination are not apparent.

Price elasticities28 for electricity demand, road fuels, and fuels in the other energy sectors are taken to be -0.5 (see Appendix 2), with half of the response coming from reduced use of energy-consuming products and half from higher energy efficiency (e.g., due to technological advances). In the power sector, the (conditional) price elasticities for fossil fuels are taken to be -0.7 based on empirical evidence and an upward adjustment to reflect the growing potential for switching to renewable generation. The same elasticities are used for all countries in the absence of systematic evidence to the contrary.29 As discussed in Appendix 2, there is a sizable (crosscountry) empirical literature to draw on for these elasticities and (as noted below) the price responsiveness of emissions is in line with that of other models.

CO2 emissions per unit of fuel use are taken from the International Energy Agency (there is little cross-country variation in emission rates per unit of energy for a given fuel product). Local air pollution mortality rates per unit of fossil fuel use are obtained from updating detailed, country-by-country estimates30 using trends in urban population exposure and declining emission rates. Policy-induced welfare impacts are calculated using applications of standard formulas (Appendix 1). These formulas do not account for linkages with the broader fiscal system and the efficiency of revenue use31 and they understate economic costs by ignoring transitory adjustment costs (e.g., idled labor and capital) and possible macroeconomic impacts from price changes, but such effects are especially challenging to model on a cross-country basis.

D. Policy Scenarios

The analysis considers a carbon tax (imposed on top of any existing fuel taxes) as well as other policies scaled so that their carbon prices (for the emissions they reduce) match those of the carbon tax—other policies, as defined here, are less effective at reducing emissions and raising revenue as the carbon tax but it is still important to have a quantitative sense of the differences (e.g., other policies may be more politically acceptable).32 A less stringent version of the carbon tax is also considered (but not for other policies).

The carbon tax is assumed to cover all fossil fuel CO2 emissions thereby promoting all opportunities for mitigating these emissions (switching to cleaner fuels, improving energy efficiency, conserving on energy-using products) across the three energy sectors. From an administrative perspective, the tax is best levied on the carbon content of fuel supply (at the point of extraction, processing/distribution, and import, ideally extending existing capacity for fuel excises or royalties).33 The carbon tax (initially zero) is assumed to rise in annual increments of $5 per ton of CO2 each year from 2017 onwards to reach $70 per ton by 2030. The ‘modest’ version of the tax rises at $2.5 per ton a year to reach $35 per ton by 2030.

In terms of global emissions coverage, ETSs are the most common form of carbon pricing.34 They are often applied, and as assumed here to differentiate them from carbon taxes, downstream on large stationary emitters, although in principle they could be extended to the supply of transportation and heating fuels (though this would essentially duplicate fuel excise

administration). The ETS is modelled implicitly by the emissions price established by the cap (set equal to the price in the carbon tax scenario), applied to the carbon content of fuels used by power generators and other large energy users;35 allowances are assumed to be fully auctioned with no earmarking of revenues (in practice many allowances, e.g. in the EU, are often given away for free).

Some G20 countries impose coal excises, but not at meaningful levels from a climate change perspective.36 A simple excise is modelled here, mimicking the coal charge portion of the broader carbon tax. Many countries also impose excises on final electricity consumption, though they are generally modest when averaged across both intermediate and final consumption and expressed in terms of CO2 equivalent.37 A tax on all electricity output is modelled here, where the tax equals the induced annual electricity price increase in the carbon tax scenario (which varies by country with the emissions intensity of generation). More effective still would be to tax electricity CO2 emissions (to promote switching to cleaner fuels) and therefore a charge on the carbon content of power generation fuels is considered (with the same CO2 price as the carbon tax). The difference between the electricity emissions tax and the electricity tax approximately indicates the effect of a policy (e.g., a CO2 per kilowatt-hour standard) to lower the carbon intensity of the power sector.

All G20 countries either tax or (in the case of Indonesia, South Africa, and Saudi Arabia) subsidize road fuels so it is useful to assess the CO2 impacts of price reform. Higher gasoline and road diesel taxes (or corresponding subsidy reductions) are considered, where the tax increases are the CO2 emissions factors for the fuels times the CO2 price in the carbon tax scenario. Sector-specific energy efficiency policies have limited environmental effectiveness relative to economywide carbon taxation and are not considered here.38 Instead, an energy efficiency combination policy is considered with a shadow price that rewards CO2 reductions from energy efficiency improvements across all sectors, leaving aside the (non-trivial) practical challenges in covering all products and equating implicit emissions prices across products and sectors.39 Finally, incentives for renewables are not modelled here due to the difficulty of obtaining clean country comparisons when barriers to scaling them up (e.g., availability of sunny/windy sites, compatibility of intermittent power) differ considerably across countries.40

III. Results

This section describes the BAU scenarios; the carbon and broader environmental, fiscal, and economic welfare impacts of carbon taxes; the relative performance of other mitigation instruments; sensitivity analyses; and (for four countries) incidence results. The focus is mostly on 2030, the target year for meeting most NDCs, to indicate potential impacts of policies when fully phased in and the extent to which they reach NDC targets.

A. BAU Scenarios

(i) CO2 and fuel mix projections

Figure 1 shows percent changes in BAU CO2 emissions by G20 country between 2015 and 2030, broken down by the contribution from changes in GDP, the ratio of primary energy to GDP, and the ratio of CO2 to primary energy. Growth in GDP (for given energy/GDP and CO2/energy ratios) would cause large CO2 increases in emerging market economies like China (140 percent), India (213 percent) and Indonesia (121 percent), while in other countries the increase is more modest (between 11 percent in Japan to 71 percent in Turkey). On the other hand, the energy intensity of GDP falls significantly, causing offsetting reductions in CO2 of between 20 percent (in Russia) to over 40 percent (Argentina, China, India, Indonesia), reflecting a combination of below-unitary income elasticities for energy products, autonomously improving energy efficiency, and the dampening demand effect of gradually rising fuel prices. By contrast, changes in the CO2 intensity of energy are modest and there is no systematic pattern to the direction, despite faster technological change in renewables (but from a low base and with no new renewables incentives). The net result of these three trends is rapid CO2 emissions increases in India (123 percent), more modest increases (between 0 and 42 percent) in nine cases, and modest (and perhaps surprising) reductions (up to 17 percent) in nine cases.

Figure 1.
Figure 1.

Change in BAU CO2 Emissions, 2015–2030

(In percent)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text and Appendix.Notes: BAU projections assume no new, or tightening of existing, CO2 mitigation polices beyond those implicit in 2014 fuel use data. The bars indicate changes in CO2 emissions from changes in GDP, the energy intensity of GDP, and the CO2 intensity of energy, while the boxes indicate the net effect.

Figure 2 shows the BAU primary fuel mix in 2030 (fuel shares are generally not that different from those in 2015 given the absence of new policies, or large changes in relative fuel prices, that would significantly shift the fuel mix). The high coal users include Australia (29 percent of primary energy), China (59 percent), India (52 percent), and South Africa (60 percent); intermediate users (with shares between 10 and 30 percent) include, Germany, Indonesia, Japan, Korea, Turkey, UK and US; and low users (with shares of 10 percent or less) include Argentina, Brazil, Canada, France, Italy, Mexico, Russia, and Saudi Arabia. Natural gas shares vary from less than 12 percent of primary energy in Brazil, France, India, and South Africa to about 30 percent or more in Argentina, Australia, Canada, Italy, Japan, Mexico, Russia, Saudi Arabia, Turkey, UK and US. Oil shares vary from 23 percent in France to over 50 percent in Brazil, Mexico, and Saudi Arabia. Nuclear accounts for about 30 percent of energy use in France, but about 10 percent or less in other G20 countries. Renewables (including biomass) represent around 5–20 percent in most cases, and about 30 percent in Brazil and 34 percent in France.

Figure 2.
Figure 2.

BAU Primary Fuel Mix, 2030

(In percent)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text and Appendix.Notes: BAU projections abstract from the possibility of enhanced policy incentives for renewables beyond 2014.

(ii) Air pollution mortality

Figure 3 shows BAU annual premature deaths in 2030 from fossil fuel outdoor air pollution.41 The death rate is highest in China, at 900 per one million of the population, reflecting China’s high coal use and high population exposure to pollution. In contrast, the coastal location of many coal plants in South Africa (where a large portion of emissions disperse offshore) and lower average population density imply a much lower mortality rate (about 50 per one million). Mortality rates from all fossil fuels are about 400 per million people in Russia, 240 in India, 185 in Korea, and 110 in Germany but below 100 per million in all other cases. In absolute terms, annual mortality is highest in China (about 680,000), followed by India (380,000), Russia (55,000), and the United States (25,000).

Figure 3.
Figure 3.

BAU Outdoor Air Pollution Death Rates from Fossil Fuels, 2030

(In units per million people)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See Appendix 2.Notes: Estimates project forward country level mortality rates per unit of fuel use in Parry and others (2014) accounting for declining air emissions rates and (for China and India) rising urban population exposure to a given level of pollution. Estimates excludes deaths from indoor air pollution and non-fossil pollution.

(iii) Emissions reductions implied by Paris pledges

Figure 4 indicates the percent reduction in fossil fuel CO2 emissions in 2030 below BAU emissions in 203042 needed to meet NDCs, assuming no tightening of mitigation policies (e.g., the EU ETS) in the BAU beyond 2014 and CO2 reduction targets proportional to those (listed in Table 1) for total GHGs.43 These reductions are substantial in many cases—over 40 percent in Australia and Canada, between 30 and 40 percent in European countries, Korea, South Africa, and United States44, between 20 and 30 percent in five cases, and less than 20 percent in Argentina, China, India, Indonesia, and Russia. This variation reflects differences in both the percent reduction targets and in the baseline year against which 2030 reductions apply: for example, for Australia, if the baseline year were 2030 rather than 2005, the required emission reduction would be 27 percent.45 For the G20 as a whole (using the 2030 BAU emissions share-weighted average), pledges amount to a reduction of about 21 percent below BAU levels in 2030.

Figure 4.
Figure 4.

Percent Reduction in BAU CO2 Emissions Implied by Paris Pledge, 2030

(In percent)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text.Notes: Estimates are the percent reduction in BAU emissions in 2030 needed to meet NDC targets, assuming fossil fuel CO2 emissions fall by the same proportion (relative to the baseline year in the Table 1 target) as implied by the percent GHG reduction targets, aside from Brazil and Indonesia where (due to the disproportionate contribution from forestry) CO2 is assumed to fall by 33 and 50 percent respectively of the percent GHG reduction. For Brazil and the United States, percent reductions in BAU emissions below 2030 levels are assumed equal to the percent reductions below BAU emissions in 2025 implied by their Paris targets.

B. Impacts of Carbon Taxes

(i) CO2 emissions

Figure 5 shows the percent reduction in CO2 emissions in 2030 (again relative to 2030 BAU CO2) induced by the two carbon tax scenarios. Under the tax rising to $70 per ton by 2030, seven countries meet or exceed their NDCs (Argentina, China, India, Indonesia, Russia, South Africa, and Turkey), another six countries would need further emissions reductions of up to 10 percent (Brazil, Germany, Japan, Korea, Mexico, Turkey, and United States), and six countries would need further emissions reductions of above 10 percent (Australia, Canada, France, Italy, Saudi Arabia, and United Kingdom) suggesting that prices well above $70 per ton would be needed in the latter cases.

Figure 5.
Figure 5.

Reduction in CO2 Emissions from Carbon Taxes, 2030

(In percent)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text.Notes: Blue and brown bars indicate percent CO2 reductions under the $35 carbon tax and additional reductions under the $70 carbon tax respectively, relative to 2030 BAU emissions. The black squares indicate the target CO2 reductions as defined in Figure 4.

This wide dispersion in required carbon prices to meet NDCs reflects both differences in their stringency (as just discussed) and in the relative price responsiveness of emissions—for example, the $70 carbon tax cuts emissions by well over 30 percent in coal-intensive China, India, and South Africa; by around 15–25 percent in Argentina, Brazil, Canada, Indonesia, Japan, Mexico, Russia, Turkey, United Kingdom and United States; and by less than 15 percent in France and Saudi Arabia (where coal use is zero or minimal).46 Also noteworthy is that emissions reductions under the $70 carbon tax are less than double those under the $35 tax, especially in coal intensive countries—in China, for example, the $35 carbon tax raises coal prices by 116 percent relative to the BAU in 2030 (cutting emissions 30 percent), but further raising it from $35 to $70 increases coal prices (relative to prices with the $35 tax) by a much smaller 54 percent (cutting emissions by a further 12 percent).

Coal accounts for more than two-thirds of the CO2 reductions in nine countries, and more than four-fifths in six cases (Australia, China, Germany, India, Korea, and South Africa), reflecting both coal’s significant share in emissions and its relatively high responsiveness to carbon pricing (e.g., the percentage increase in coal prices is typically around 10–15 times the percentage increase in road fuel prices). In contrast, reductions in oil use account for about 30 percent or less of CO2 reductions in all but three cases (Brazil, France and Saudi Arabia). Natural gas accounts for 30 percent or less of CO2 emissions reductions in all but six cases (Argentina, Canada, Italy, Mexico, Russia, and Saudi Arabia).

A final noteworthy point from Figure 5 is that, at the G20 level, even the $35 CO2 price is sufficient to meet the (2030 BAU emissions share-weighted average) Paris pledge (given the disproportionate weight attached to the more modest mitigation pledges of China and India).

(ii) Revenue

Figure 6 shows revenue from the carbon tax in 2030. Revenues are potentially large, typically around 1–2.5 percent of GDP for the $70 tax, and substantially higher in a few countries with high CO2 intensity of GDP including India, Russia, Saudi Arabia, and South Africa. Revenues are about 70–85 percent higher under the $70 per ton tax compared with the $35 tax (they are less than double due to the reduction in fuel demand). For a given tax rate (though not shown in the figure), revenues as a percent of GDP decline gradually over time with the declining emissions intensity of GDP—for example, a $70 tax in China in 2020 would raise revenues of 3.7 percent of GDP compared with 3.2 percent in 2030.47

Figure 6.
Figure 6.

Revenue from Carbon Taxes, 2030

(In percent of GDP)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text.Notes: Calculations account for induced changes in revenues from pre-existing fuel taxes but not from the broader fiscal system (see text footnote).

(iii) Health

Figure 7 indicates the percent reduction in fossil fuel air pollution deaths below BAU levels in 2030 from carbon taxes. Comparing with Figure 5, the percent reductions are moderately larger than the percent CO2 reductions (for the corresponding carbon tax) in some cases where coal accounts for a larger share of economy-wide air pollution deaths than for economywide CO2 emissions. More noteworthy perhaps are the considerable differences in average lives saved per 1 million tons of CO2 reduced, which are about 5–20 in nine cases, 4 or less in Argentina, Australia, Canada, Mexico, and Saudi Arabia, but about 50 and 80 in China and India respectively. If, for illustration, the value of a statistical life for China is $1 million48 the domestic environmental benefit per ton of CO2 reduced in China (from reduced local air pollution not the reduced CO2 itself) would be about $100.

Figure 7.
Figure 7.

Reductions in Air Pollution Deaths from Carbon Taxes, 2030

(In percent)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source. See text.

(iv) Economic welfare gains

Figure 8 shows the economic welfare costs (i.e., losses in consumer surplus less government revenue gains) of carbon taxes and their net domestic welfare gains (i.e., domestic environmental benefits—excluding global climate benefits—less welfare costs) in 2030 as a percent of GDP. Economic costs are less than 0.5 percent of GDP under the modest tax, and less than 0.8 percent of GDP under the $70 tax in all but three cases—China, India, and South Africa where costs are 1.0–1.6 percent of GDP. However, for the modest carbon tax, domestic environmental benefits are about as large as, or larger than, these costs in all cases (mostly due to domestic air pollution benefits), so net welfare gains are zero or positive. Even for the $70 tax, although there are net welfare losses in three cases (in two cases these losses are only 0.1 percent of GDP, while in South Africa they amount to 0.7 percent of GDP), there are large net welfare gains of: 0.7 percent of GDP in Korea, 2.3 percent in India, 3.7 percent in Russia, and 6.7 percent in China. These results suggest that many G20 countries can move ahead unilaterally with carbon pricing and perhaps make themselves better off in net terms or incur only small net costs49 (for reasons noted earlier economic costs may be understated).

Figure 8.
Figure 8.

Domestic Welfare Effect of Carbon Taxes, 2030

(In percent of GDP)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text.Note: Environmental benefits include reduced domestic air pollution mortality, traffic congestion, traffic accidents, and road damage, but exclude global climate benefits. Welfare costs reflect losses in consumer surplus in fuel markets from carbon charges net of revenue gains to the government accounting for any erosion of revenues from pre-existing fuel taxes (there are no changes in producer surplus given perfectly elastic supply curves). The net welfare gain is the domestic environmental benefit less the economic cost.

C. Comparison of Alternative Mitigation Policies

Table 2 shows the impacts of other mitigation instruments, relative to the $70 carbon tax, for CO2 reductions, revenue, and reductions in air pollution deaths (first three quadrants)50 as well as absolute welfare impacts (as a percent of GDP) in the fourth quadrant.

The coal tax is quite effective in some cases, achieving around three quarters or more of the reductions under the $70 carbon tax in Australia, China, Germany, India, Korea, and South Africa, and over 60 percent of the CO2 reductions in Indonesia, Japan, and Turkey. The ETS is significantly less effective than the carbon tax (not because of the instrument itself but rather its assumed coverage of large emissions sources only)—it reduces emissions by between about 40 and 75 percent of the reductions under the carbon tax in 16 cases, only 27 percent in France, and about 80 percent or more in Australia and India. In fact, in 10 cases the coal tax is about as effective, or more effective, than the ETS—only in Argentina, Canada, Italy, Mexico, Russia, and Saudi Arabia is the ETS substantially more effective. The electricity (CO2) emissions tax achieves 80 percent or more of the CO2 reductions under the ETS in all but two cases. In general, the electricity output tax is substantially less effective than the electricity emissions tax (as it does not promote switching among power generation fuels), though in a few cases (e.g., South Africa with limited possibilities for fuel switching) the difference is relatively modest. Road fuel taxes have weak effectiveness, well below 10 percent that of the carbon tax for all but one case (given their sectoral focus and the relatively modest proportionate impact of carbon charges on road fuel prices). The energy efficiency combination policy is more effective, typically reducing emissions by between 25 and 40 percent that of the carbon tax (though as noted above there are practical obstacles to its comprehensive, cost-effective implementation).

Coal taxes are relatively less effective at raising revenue than reducing CO2 emissions—in 16 cases coal tax revenues are around 30 percent or less of carbon tax revenues, given that coal accounts for a minor fraction of total emissions and its disproportionately high responsiveness to carbon pricing (the notable exceptions are China, India, and South Africa where revenues are 63–76 percent of carbon tax revenue). The revenue potential of ETSs is generally around 20–60 percent that of the carbon tax, primarily because ETSs tend to miss about half of the potential revenue base (and the portion they do cover is typically more mobile due to relatively low-cost mitigation opportunities in the power sector). Generally, revenues from taxes on electricity, or electricity emissions, are moderately less than those from ETSs (as they do not tax emissions from large industrial firms). On the other hand, road fuel taxes are relatively more effective at raising revenues than they are at reducing CO2 emissions, raising around 15–30 percent of the revenues under the carbon tax. The energy efficiency combination slightly erodes bases for pre-existing fuel taxes, but the revenue losses are very small.

The performance of other carbon mitigation policies at reducing air pollution deaths, relative to the carbon tax, follows a broadly similar pattern to that of the relative reductions in CO2 emissions, though with some nuances. For example, in some cases, the relative performance of the coal tax is better.

Welfare results can be largely anticipated by the previous discussion. For example, in India the welfare gains from the carbon tax, coal tax, ETS, electricity output tax, electricity CO2 tax, road fuel tax, and energy efficiency combination are 2.3, 1.8, 1.5, 0.5, 1.3, 0.3 and 0.8 percent of GDP respectively. In many other cases, welfare impacts are far less striking. For example, in the United States the carbon tax, road fuel tax, and energy efficiency combination generate modest net benefits (0.04–0.06 percent of GDP) while the other policies generate modest net costs (0.02–0.05 percent of GDP). However, only the carbon tax comes close to meeting the Paris pledge.

D. Sensitivity Analysis

The relative impacts of other mitigation policies are either approximately robust to alternative parameter assumptions, or affected in largely predictable ways (e.g., increasing road fuel price elasticities increases the environmental effectiveness of road fuel taxes in rough proportion). The discussion here, therefore, focuses on the emissions impacts of carbon pricing and how it compares with the Paris mitigation targets, varying the main uncertain parameters by plus and minus 50 percent (which, loosely speaking, is in line with parameter ranges discussed in Appendix 2).

In cases (Argentina, Indonesia, Korea, Mexico, and Turkey) where Paris pledges specify percent reductions below BAU levels in 2030 (Table 1), changes in absolute 2030 BAU emissions, due to parameter variations, do not affect the required percent reductions in 2030. In other cases, where Paris pledges are relative to historical emissions, increases in 2030 BAU emissions will increase the required percent reduction in 2030 and vice versa for reductions in BAU emissions. In this regard, changes in GDP growth rates, income elasticities, and fuel price elasticities all noticeably affect the required percent reductions in 2030, while changes in the rates of autonomous technological change and international energy prices have a smaller effect. Required percent CO2 reductions in China (and to a lesser extent India) are especially sensitive to GDP and income elasticity assumptions—for example, higher GDP growth, combined with the baseline assumption of low income elasticities for energy products, lowers the emissions intensity of GDP and China’s Paris pledge is met in the BAU.

Percent reductions in CO2 below 2030 BAU levels induced by the $70 carbon tax are not very sensitive to assumptions about GDP growth, income elasticities, or autonomous rates of technological change.51 They are, however, sensitive to fuel price elasticities: percent CO2 reductions fall by about two-fifths when elasticities are 50 percent smaller in size than in the baseline, and increase by around one-third when elasticities are 50 percent larger. They are also almost as sensitive to different scenarios for international energy prices, as these affect the proportionate increase in prices from BAU levels induced by carbon charges.

E. Incidence Analyses

Standard procedures involving input-output tables and household expenditure surveys, as described in Appendix 3, are used to obtain first order approximations of incidence effects (i.e., abstracting from behavioral responses) for carbon taxes in four countries, assuming full pass-through of taxes into consumer prices in domestic markets.52

The burden of carbon taxation on households and firms is discussed below for Canada, China, India, and the United States, focusing on year 202053 (since these impacts are of immediate concern as the tax is phased in), and a tax of $20 per ton of CO2.54

(i) Households

Figure 9 shows the burdens as a percent of total expenditure for household consumption quintiles. Burdens are defined relative to consumption rather than income as the former is generally viewed as a better proxy for ‘permanent’ or lifetime income.55 There are several noteworthy points.

Figure 9.
Figure 9.

Burden of $20 Carbon Tax on Household Consumption Quintiles, 2020

(In percent of total household consumption)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

Source: See text.Note: Quintiles 1 and 5 are the lowest and highest consumption groups respectively.

First, the carbon tax is moderately regressive in China, significantly more regressive in the United States, mildly progressive in India, and initially progressive then regressive in Canada. Burdens on the bottom and top income deciles are 4.5 and 3.3 percent of consumption respectively in China, 2.8 and 1.4 percent respectively in the United States, 1.9 and 2.6 percent respectively in India, and 1.1 and 1.3 percent respectively in Canada.56

Second, in China, India, and the United States the regressivity or progressivity of the tax is largely driven by differing budget shares for electricity—declining for higher consumption households in China and the United States, and rising for such households (partly reflecting higher rates of grid access) in India. The indirect burdens from the higher prices of consumer goods in general (induced by higher energy prices) are larger in China and India than in Canada and the United States (reflecting, in the former, the greater dependence of consumption on domestically manufactured products whose prices rise from carbon pricing)—and are approximately proportional to expenditure across household quintiles (which significantly moderates the overall regressivity or progressivity of the carbon tax).

Third, the fiscal costs of compensating the bottom income quintile for these burdens are not too large (as most of the burden of higher prices is borne by higher income groups): about 11 percent of the carbon tax revenues in Canada, 13 percent in the United States, 5 percent in India, and 8 percent in China.57 Full compensation would nonetheless become increasingly challenging as the carbon price is progressively ramped up over the longer term.58

(ii) Industry

Figure 10 indicates the average percent cost increases for most vulnerable (i.e. those incurring the highest cost increases) exporting industries (weighted by each industry’s share in total exports)59 for the same countries and $20 per ton carbon tax in 2020. For instance, in India the 20 percent of most vulnerable exporters face weighted average cost increases of 5.4 percent (the largest increase is 10 percent for non-ferrous basic metal exports which account for 1.8 percent of total exports), the 40 percent of most vulnerable exporters face weighted average cost increases of 3.8 percent, and across all exports the weighted average cost increase is 2.2 percent. Looking across countries, cost increases are largest for China (with a weighted average increase across all exporters of 8.2 percent) reflecting the high energy-intensity of exports and carbon intensity of energy, and smallest for Canada (with a weighted average cost increase across all exporters of 0.7 percent) given its limited use of coal.

Figure 10.
Figure 10.

Cost Increases from $20 Carbon Tax by Exporter Quintile, 2020

(In percent of total costs)

Citation: IMF Working Papers 2018, 193; 10.5089/9781484373842.001.A001

IV. Conclusion

One robust theme from the above quantitative analysis is the substantial environmental, fiscal, and welfare advantage of (comprehensive) carbon taxes (or equivalent instruments) over other mitigation policies across G20 countries. Such taxes can be in many countries’ own national interests when accounting for the domestic environmental benefits (before even counting climate benefits). This raises the issue of how to move carbon pricing forward at the domestic level.60

Previous experience with energy price reform61 suggests a mix of ingredients may be needed to advance domestic carbon pricing. One is to have a comprehensive plan addressing stakeholder concerns (e.g., from legislators, industry, consumer groups, unions), with clearly stated objectives, timetables, and specifics (e.g., on revenue use). Another ingredient is an effective communications plan informing the public about both the global and the national (environmental, health, and fiscal) benefits of reform. Gradual reforms providing firms and households time to adjust are also desirable. The biggest challenge consists of addressing burdens on vulnerable groups, which requires: improved targeting of social safety nets; displaced worker programs;62 and possible temporary tax reliefs for energy-intensive industries (though on efficiency grounds these should be progressively phased out, especially if other countries are acting on their Paris pledges).

Another robust theme is the high carbon prices—at least given evidence on fuel price responsiveness and the assumption of no other mitigation policies—that might be needed to meet NDCs in some cases. This might imply increases in energy prices (and burdens on vulnerable groups) that push the bounds of political acceptability. There is also a need to complement carbon pricing with other policies to enhance its environmental effectiveness, such as infrastructure upgrades to accommodate renewable generation, carbon capture and storage, electric vehicles and so on.

A third theme is the potentially wide cross-country dispersion in prices needed to implement mitigation pledges, suggesting opportunities for improving cost-effectiveness through some degree of price coordination at the international level. A relative tightening of lax NDCs would promote some carbon price convergence. An explicit carbon price floor arrangement (presumably among large emitters and as a reinforcement—not substitute—for the NDC process) could also help. Such an arrangement (similar to carbon price floor requirements at the provincial/territorial level in Canada63 and, more generally, for indirect taxes incorporated into national legislation for EU member states) provides some degree of protection against competitiveness concerns, allows countries flexibility to set prices higher than the floor if needed for their NDCs, and could benefit all participants.64

The type of model developed here, with consistent and transparent comparisons across policies, metrics, countries, and parameter scenarios, has a useful role in informing dialogue on the domestic and international design of carbon pricing and related mitigation policies (subject to the caveats and uncertainties detailed above).

Table 3.

Sensitivity of CO2 Reductions from Paris Pledge and $70 Carbon Tax, 2030

article image
Source: See text and Appendix.Note: Due to rounding, in some cases slight changes in baseline CO2 reductions due to parameter variations are not apparent.

Appendix 1. Analytical Model

A discrete time-period model is used where t=0...t¯ denotes a year (though most focus is on 2030). Fossil fuels are first discussed, followed by fuel use in the power, road transport, and other energy sectors.

(i) Fossil Fuels

Coal, natural gas, gasoline, road diesel and other oil products are denoted by i = COAL, NGAS, GAS, DIES, and OIL respectively. The user fuel price at time t, denoted pti, is:


τti is the tax or subsidy (if negative) on fuel i reflecting (a) the combined effect of any pre-existing excises, favorable treatment under general sales tax for household fuels, and distortions from regulated or monopoly pricing and (b) any carbon charge. For most countries, τti is large for road fuels and zero for coal and gas (or a small positive for the latter two fuels in countries covered by the EU ETS). p^ti is the pre-tax fuel price or supply cost (the international commodity price adjusted for processing/distribution margins). For fuel products used in multiple sectors, pre-tax prices and taxes are taken to be the same for all fuel users (for non-road oil products taxes are zero, except for EU countries to the extent they are covered by the ETS).

(ii) Power Sector

Residential, commercial, and industrial electricity consumption is aggregated into one economy-wide demand for electricity in year t, denoted YtE, and determined by:


UtE is usage of electricity-consuming products or capital (i.e., the stock of electricity-using capital times its average intensity of use). htE is the electricity consumption rate (e.g., kWh per unit of capital usage), the inverse of energy efficiency. Product use increases with gross domestic product (GDPt)according to vE, the (constant) income elasticity of demand for electricity-using products. Product use also varies inversely with proportionate changes in unit electricity costs—the electricity consumption rate times the user electricity price ptE. ηUE < 0 is the (constant) elasticity of demand for use of electricity-consuming products with respect to unit energy costs.65 The electricity consumption rate declines (given other factors) at a fixed annual rate of αE ≥ 0, reflecting autonomous energy efficiency improvements (e.g., due to gradual retirement of older, less efficient capital). Higher electricity prices further increase energy efficiency, implicitly through adoption of more efficient technologies: ηhE is the elasticity of the energy consumption rate with respect to energy prices. Note that (A2) can be implemented with U0Eandh0E normalized to unity (absolute values for these parameters are not needed).

Power generation fuels potentially include coal, natural gas, oil, nuclear, hydro, biomass and (non-hydro, non-biomass) renewables (principally wind and solar), where the latter are denoted by i = NUC, HYD, BIO, and REN. To accommodate flexible assumptions for the degree of substitution among fuels, the share of fuel i in generation, denoted θtEi, is defined:


where i,j,l = COAL, NGAS, OIL, NUC, HYD, BIO, REN. gti is the full cost of generating a unit of electricity using fuel i (fuel, labor, capital, transmission/distribution costs). ϵ˜Ei<0 is the conditional (indicated by ~) own-price elasticity of generation from fuel i with respect to generation cost. Conditional here means the elasticity reflects the percent reduction in use of fuel i due to switching from that fuel to other generation fuels, per one-percent increase in generation cost for fuel i, holding total electricity generation fixed. Generation cost elasticities are larger than corresponding fuel price elasticities as an incremental increase in fuel and non-fuel generation costs has a bigger impact than an incremental increase in fuel costs alone.

From (A3) fuel i’s generation share decreases in own generation cost. It also increases in the generation cost of fuel; j ≠ i, where the increase in fuel i’s generation share is the reduced share for fuel; (i.e., θ0Ej times the term in square brackets) multiplied by the (initial) share of fuel/in generation from all fuel alternatives to; (i.e., θ0Ei/Σljθ0El).

Use of fossil fuel/in power generation at time t, denoted FtEi, is given by:


Fuel use equals the generation share times total electricity output—assumed equal to total electricity demand—and divided by ρtEi, the productivity of fuel use or electricity generated per unit FtEi.

Unit generation costs are determined by:


where ktEi is non-fossil fuel costs per unit. Unit generation costs decline with rising productivity (which, for a given generation type, is assumed to reduce fossil-fuel and non-fuel costs by the same proportion). Productivity of generation by fuel i increases at rate αpi ≥ 0 per year (again, for example, due to retirement of older, less efficient plants).



The user price of electricity is the generation shares times unit generation costs summed over fuels (pre-existing electricity taxes are taken to be zero

(iii) Road Transport Sector

Analogous to (A1), gasoline and road diesel fuel demand in period t, denoted FtTi, where i = GAS, DIES is gasoline and diesel respectively, is:


UtTi is kilometers (km) driven by vehicles with fuel type i and htTi is average fuel use per vehicle km (the inverse of fuel economy). km driven in vehicle type i increases with GDP, according to the income elasticity of demand vTi, and varies inversely with proportionate changes in fuel costs per km htTipti where ηUTi < 0 is the elasticity of vehicle km driven with respect to per km fuel costs.66 αTi ≥ 0 is an annual reduction in the fuel consumption rate due to autonomous technological change improving fuel economy. Higher fuel prices also reduce fuel consumption rates according to the elasticity of the fuel consumption rate ηhTi ≤ 0—this encompasses both improvements in petroleum vehicles (better engine efficiency, lighter weight materials, shifting to smaller vehicles, etc.) as well as shifting to electric and hybrid vehicles.

(iv) Other Energy Sector

The other energy sector is decomposed into large and small energy users, the latter representing households and small firms (with emissions below a threshold), denoted by q = LARGE, SMALL, respectively. Use of fuel i in the other energy sector, by group q, at time t, denoted FtOqi, is:


where i = COAL, NGAS, OIL, BIO, and REN. The interpretation for (A8) is analogous to that for (A2) and (A7) with UtOqiandhtOqi denoting respectively, use of products requiring fuel i at time t by group q and its fuel consumption rate. Parameters vOi, ηUOi, ηhOi, and αOi have analogous interpretations to previous notation and are taken to be the same across large and small users.

(v) Modelling Policies

The carbon tax is modelled by incorporating into the tax for fuel i a charge of τtCO2μCO2i, for i = COAL, NGAS, GAS, DIES, and OIL, where τtCO2 is a uniform tax on CO2 emissions in period t and μCO2i is fuel i’s CO2 emissions factor (positive for fossil fuels and zero for renewables, hydro, biomass, and nuclear).67 The ETS is modelled in the same way, but with charges applying only to fuels used by power generators and large users in the other energy sector. The coal tax is the same policy as the carbon tax, but with charges applying only to coal use, and similarly the road fuel tax applies the carbon charges to road fuels only. The electricity tax, denoted τtE, increases electricity prices by the same amount as they increase in the carbon tax scenario. The electricity emissions tax is the same policy as the ETS but with charges applied to power generation fuels only (not large industrial users). The energy efficiency policy applies a ‘virtual’ carbon charge to fuel prices in the equations governing energy efficiency, but not to fuel prices in the equations governing the usage of energy-consuming products.68

(vi) Metrics for Comparing Policies

CO2 emissions. CO2 emissions from fossil fuel use at time t are:


where j = E, T, O denotes the electricity, road transport and other energy sector respectively.

Revenue. Revenue from fuel and electricity taxes is:


Deaths from fossil fuel air pollution. At time t these are given by:


mtji is mortality per unit of (fossil) fuel i used in sector j, which may differ by sector due to differing use of air emissions control technologies and local population exposure to emissions.

Economic welfare gains. The economic costs and net welfare gains of policies are calculated using applications and extensions of long-established formulas in the public finance literature69 based, for simplicity, on second order approximations.70 The information required to apply these formulas includes the size of price distortions in fuel and electricity markets (i.e., the difference between social and private costs due to any domestic environmental costs net of any fuel taxes), any induced quantity changes in markets affected by these distortions (an output from the model), and any new source of distortions created by carbon policies.71

The economic welfare gains (excluding the global climate benefits) from a carbon tax in period t are computed using:


where a ∧ denotes a BAU value and Γtji is the price distortion in a fuel market.

In (A13), Γtji consists, for non-road fossil fuels, of local air pollution costs, equal to premature mortalities per unit of fuel use times VMORTt, the value per premature mortality—Γtji is defined net of any pre-existing fuel taxes but these are modest at most for non-road fuels. For road fuels, there is an additional environmental cost equal to the external costs of traffic congestion, accidents, and road damage expressed per unit of fuel use, βtTi. The latter is multiplied by the term in parentheses, which amounts to the fraction of the induced change in fuel use due to changes in vehicle km driven as opposed to (long run) improvements in fuel economy.72

In (A14), ΔFtji is the change in fuel use, relative to its BAU level F^tji.

From equation (A12), the welfare change from the tax increase for a fossil fuel in a sector consists of: (i) the reduction in use of a fuel product in a particular sector times the pre-existing price distortion associated with that product/sector and aggregated over fuels/sectors, less (ii) the ‘Harberger triangle’, equal to the reduction in fuel use times one-half of the tax increase, where the latter is the product of the fuel’s CO2 emissions factor and the CO2 price at time t.

The above formula is also used to calculate the net welfare change from the ETS, coal tax, road fuel taxes, and energy efficiency policies, where the charges apply to fuel use (or virtually to energy efficiency) and sectors as described above.

Appendix 2. Model Parametrization

The model pivots off fuel use data for 2014. Data for 2015 is available, but does not fully reflect the sharp drop in energy prices between 2014 and 2015 (i.e., 2015 fuel data may significantly understate current equilibrium fossil fuel use).

(i) Fossil Fuels

Supply prices for coal, natural gas, gasoline, diesel, and oil products for 2014–2015 by country are from the IMF 73 and reflect international reference prices of the finished product (e.g., gasoline) adjusted, where appropriate, using standard (absolute) markups for transport and distribution costs.74 The international (crude) component of these prices is projected forward using actual and projected international energy prices obtained by averaging over projections in IMF (2017)75 and EIA (2018), Tables 12, 13 and 15.76 The IMF projections suggest coal prices increase 30 percent between 2015 and 2030, oil prices stay the same,77 and North American gas prices change little, though prices for European gas and liquefied natural gas decline. EIA (2018) projects an 86 percent increase in oil prices, a 32 percent increase in coal prices, and gas price increases of 23–66 percent (Table Appendix 1). BAU CO2 emissions, and CO2 prices needed to induce a given percent reduction in emissions, are generally higher when future international energy prices are assumed to be lower.

Table 1.

Comparison of Future Fuel Price Assumptions

article image
Note: All prices in $2015. ‘2015 (Actual)’ prices are from IMF (2017).

For electricity (generally a non-traded good), the supply cost for 2014–2015 in the IMF database is the domestic production cost or cost-recovery price, with costs evaluated at international reference prices. Electricity prices are projected forward using (A6), and changes in fuel prices and generation shares in a future year relative to 2014 levels.

For G20 countries, 2014–2015 prices to fuel users are available from the IMF and the difference between these prices and supply prices, after the latter (for household fuels) have been marked up for GST, gives the estimated fuel tax (or subsidy). Pre-existing fuel taxes are taken as constant (in real terms) from 2016 onwards in the BAU while (given recent attempts to liberalize domestic energy prices in many energy-exporting countries) any subsidies (primarily for natural gas in Argentina and road fuels in Saudi Arabia) are assumed to phase out progressively by 2030.

(ii) Power Sector Electricity Consumption.

This is obtained for 2014 from IEA (2017c) focusing on domestic generation (i.e., including exported generation where fuels are combusted domestically, but not imported generation), as this is what matters for a country’s CO2 emissions.

Income elasticity of demand for electricity-using products. Empirical studies for different countries suggest a range for this elasticity of around 0.5–1.5.78 For all but two G20 countries we use a baseline value of 0.75, and for China and India values of 0.5 and 0.9 respectively, which (along with other assumptions) imply trends in the ratio of electricity consumption to GDP that are broadly consistent with projections in IEA (2017) when we use their energy price projections. The lower value for China makes an adjustment for the ongoing rebalancing of the economy away from energy-intensive industries to services79, while a higher value in India makes an adjustment for the expected progressive expansion in grid access among lower income households.

Price elasticities for electricity. A simple average across the 26 estimates of long-run electricity demand elasticities reported for different countries in Jamil and Ahmad (2011), Table 1, is about -0.5, and nearly all estimates lie within a range of about -0.15 to -1.0.80 A study for China suggests an elasticity of -0.35 to -0.5.81 Evidence for the United States suggests the long-run price elasticity for electricity demand is around -0.4, with about half the response reflecting reduced use of electricity-consuming products and half improvements in energy efficiency.82 Values of -0.25 are assumed for both the usage and energy consumption rate elasticities for all countries, implying a total electricity demand elasticity of -0.5.

Annual rate of efficiency improvement for electricity-using products. This parameter (which is of moderate significance for the BAU) is taken to be 1 percent a year.83

Generation shares. These are obtained from IEA (2017c) by the electricity produced from each fuel type divided by total electricity generation.

Own-price elasticities for generation fuels (conditional on total electricity output). Empirical studies tend to suggest that coal is only moderately price responsive. For example, one survey of eight studies for various advanced countries, China, and India put the coal price elasticity at -0.15 to -0.6.84 And for the United States, simulations from a variant of the US Department of Energy’s National Energy Modeling System (NEMS) model suggested a coal price elasticity of around -0.15 (with fuel switching rather than reduced electricity demand accounting for over 80 percent of the response).85 Other studies suggest somewhat larger responsiveness however, for example, EIA (2014) estimate a $34 per ton carbon tax raising coal prices by about 150 percent reduces US coal use by 32 percent in 2040, while an $85 per ton carbon tax reduces coal use 90 percent).86 And a study for China reports coal price elasticities of -0.3 to -0.7.87 Our judgment is that the rapid (and continued future) decline in the costs of renewable energy will likely increase the price responsiveness of coal use relative to previous estimates, and could induce significant technological innovation88, and a coal price elasticity of -0.7 is assumed here for all countries.89 The same elasticity is assumed for other fossil generation fuels. The elasticities in equation (A3) are defined with respect to generation costs rather than fuel costs and can be obtained by dividing the fuel price elasticity by the share of fuel costs in generation costs, which for coal generation is taken to be 0.25 (see below).

Fossil fuel consumption and productivity. Consumption of coal, natural gas, and oil used in power generation is taken from IEA (2017c) for 2014. Electricity generated from a fossil fuel, divided by input of that fuel, gives the productivity of the fuel.

Annual rate of autonomous productivity improvement. Productivity improvements at power plants reflect improvements in technical efficiency and gradual retirement of older, less efficient plants. For coal, annual average productivity growth is taken to be 0.5 percent based approximately on IEA (2015), Figure 2.16. For natural gas, nuclear, and hydro, there is likely a bit more room for productivity improvements and baseline annual growth rate of 1 percent is assumed. For renewables, a productivity growth rate of 5 percent is used (i.e., costs halve every 15 years).

Non-fuel generation costs. For coal and oil plants, non-fuel generat